Balancing Urban Growth and Food Autonomy: An Integrated Machine Learning and Agricultural Statistics Framework for Local Rice Self-Sufficiency in the PNAR of Purwokerto, Indonesia

Rapid peri-urbanization intensifies competition between settlement growth and farmland, creating structural risks to local food security. This study integrates machine learning–based spatial modeling, agricultural statistics, and policy-relevant scenarios to examine how land use trajectories influence local rice self-sufficiency in the Proposed New Autonomous Region (PNAR) of Purwokerto, Indonesia. Multi-temporal Sentinel-2 imagery served as input for Random Forest–based classification of existing land use, whereas CA–ANN was used to simulate transitions and predict land use in 2029. The Random Forest classification model achieved an overall accuracy of 84% with a Kappa coefficient of 0.81, while CA–ANN model validation through hindcasting for 2015–2024 reached 82% spatial agreement, with strong class stability for paddy fields (0.853) and built-up areas (0.973). Under the business-as-usual path, paddy fields decline from 8,506 ha (2015) to 6,512 ha (2029), shifting the rice balance from a +3,791 tons surplus to a –12,520 tons deficit. A combined scenario cropping index of 250, a 2% conversion reduction, and 4% population moderation restores near-equilibrium (+190 tons). These findings confirm that safeguarding peri-urban food resilience requires coordinated land use regulation, demographic control, and agrotechnological intensification. The validated CA–ANN framework offers a transferable decision-support tool for sustainable land-food governance in rapidly growing regions of the Global South.